Towards trustworthy multi‐modal motion prediction: Holistic evaluation and interpretability of outputs

Predicting the motion of other road agents enables autonomous vehicles to perform safe and efficient path planning. This task is very complex, as behaviour depends on many factors number possible future trajectories can be considerable (multi-modal). Most prior approaches proposed address multi-modal prediction are based complex machine learning systems that have limited interpretability. Moreover, metrics used in current benchmarks do not evaluate all aspects problem, such diversity admissibility output. The authors aim advance towards design trustworthy systems, some requirements for Trustworthy Artificial Intelligence. focus evaluation criteria, robustness, interpretability outputs. First, comprehensively analysed, main gaps identified, a new holistic framework proposed. Then, method assessment spatial temporal robustness introduced by simulating noise perception system. To enhance outputs generate more balanced results framework, an intent layer attached models effectiveness this approach assessed through survey explores different elements visualisation intentions. findings make significant contribution development vehicles, advancing field greater safety reliability.